Colloquium: “Simulating and Understanding Point Defects using Density Functional Theory and Graph Neural Networks” with Professor Arun Mannodi Kanakkithodi (Purdue University)
Colloquium
March 4, 2026
3:00 PM - 4:00 PM
Location
238 2SES
Calendar
Download iCal FileAbstract:
DFT is routinely used to simulate point defects in solids and calculate their formation energies as a function of chemical growth conditions, Fermi level, and defect charge. Defect energy plots help ascertain the donor- and acceptor-type nature of defects, their relative stabilities, their shallow or deep levels, the equilibrium Fermi level, and temperature-dependent defect concentrations [1,2]. The need for large supercells, charge states, and advanced functionals makes defect calculations very expensive, prohibiting their application to massive semiconductor-defect chemical spaces. Combining DFT with machine learning (ML) helps address the computational expense by enabling on-demand predictions of defect energetics and defect levels directly from descriptor-based or structure-based representations. In this seminar, I will discuss my group’s work in combining high-throughput DFT computations with crystal graph neural network (GNN) models for understanding point defect behavior in a variety of chalcogenide and halide semiconductors [3,5,6]. We developed a computational workflow that uses both semi-local and hybrid functionals to generate datasets of native point defects, impurities, dopants, and defect complexes, also accounting for energy-lowering symmetry-broken configurations [6]. GNN-based interatomic potentials trained on this data subsequently enable prediction and optimization of thousands of new point defects and complexes, and identification of the lowest energy defects. This scheme was applied for rational discovery and screening of low energy defect structures in dozens of semiconductors belonging to: (a) Cd/Zn-Te/Se/S compositions, relevant for CdTe solar cells, (b) a variety of inorganic halide (e.g., CsPbI3) and chalcogenide (e.g., BaZrS3) perovskites, and (c) zincblende-derived ternary and quaternary chalcogenides (e.g., Cu(In,Ga)S2 and Ag2ZnSnSe4).
References
[1] M.H. Rahman et al., J. Phys. Mater. 8 022001 (2025).
[2] A. Mannodi-Kanakkithodi et al., Patterns. 3, 3, 100450 (2022).
[3] M.H. Rahman et al., APL Machine Learning. 2, 016122 (2024).
[4] M. Biswas et al., J. Chem. Inf. Model., 66 (3), 1353-1370 (2026).
[5] M.H. Rahman et al., under review, preprint: https://arxiv.org/abs/2510.23514 (2026).
[6] I. Mosquera-Lois et al., npj Comput. Mater. 9, 25 (2023).
Date posted
Feb 26, 2026
Date updated
Feb 26, 2026